
import os
import seaborn as sns
from pathlib import Path
import matplotlib.pyplot as plt

sns.set(palette="Set2")


!git clone https://github.com/nazianafis/Neural-Style-Transfer
# Перейдем в директорию скачанного репозитория
%cd /content/Neural-Style-Transfer/

# Небольшая обработка скрипта, не менять!
file_path = "NST.py"
with open(file_path, "r") as file:
    # Убираем автоматический запуск
    lines = file.readlines()[:183]
# Вывод логов только каждые 100 эпох
lines = lines[:175] + ["            if cnt % 100 == 0:\n"] + lines[175:]
lines[176] = "    " + lines[176]
with open(file_path, "w") as file:
    file.writelines(lines)

from NST import *

Cloning into 'Neural-Style-Transfer'...
remote: Enumerating objects: 379, done.
remote: Counting objects: 100% (154/154), done.
remote: Compressing objects: 100% (119/119), done.
remote: Total 379 (delta 72), reused 105 (delta 34), pack-reused 225
Receiving objects: 100% (379/379), 92.21 MiB | 14.34 MiB/s, done.
Resolving deltas: 100% (143/143), done.
/content/Neural-Style-Transfer


def run_neural_style_transfer(content_image="c1.jpg", style_image="s1.jpg"):
    """
    Функция для запуска метода Neural Style Transfer.

    :param content_image: имя файла картинки контента
    :param style_image: имя файла картинки стиля
    :return: results_path -- путь к папке, в которую сохранился результат
    """
    PATH = ""
    default_resource_dir = os.path.join(PATH, "data")
    content_images_dir = os.path.join(default_resource_dir, "content-images")
    style_images_dir = os.path.join(default_resource_dir, "style-images")
    output_img_dir = os.path.join(default_resource_dir, "output-images")
    img_format = (4, ".jpg")

    # Задаем конфиги, включая вес компонент лосса
    optimization_config = {
        "content_img_name": content_image,
        "style_img_name": style_image,
        "height": 400,
        "content_weight": 100000.0,
        "style_weight": 30000.0,
        "tv_weight": 1.0,
    }
    optimization_config["content_images_dir"] = content_images_dir
    optimization_config["style_images_dir"] = style_images_dir
    optimization_config["output_img_dir"] = output_img_dir
    optimization_config["img_format"] = img_format

    results_path = neural_style_transfer(optimization_config)
    return results_path


def visualize_style_transfer(content_image, style_image, results_path):
    """
    Визуализация результатов.

    :param content_image: имя файла картинки контента
    :param style_image: имя файла картинки стиля
    :param results_path: путь к папке, в которую сохранился результат
    """
    if len(os.listdir(results_path)) == 0:
        return
    plt.figure(figsize=(13, 3))
    folders = [
        "content-images/",
        "output-images/" + Path(results_path).name + "/",
        "style-images/",
    ]
    combined_image = (
        Path(content_image).stem + "_" + Path(style_image).stem + ".jpg"
    )
    files = [content_image, combined_image, style_image]
    titles = ["Контент", "Результат", "Стиль"]
    for i in range(3):
        plt.subplot(1, 3, i + 1)
        plt.imshow(plt.imread("data/" + folders[i] + files[i]))
        plt.title(titles[i])
        plt.xticks([])
        plt.yticks([])


content_images = sorted(os.listdir("data/content-images"))
style_images = sorted(os.listdir("data/style-images"))

result_paths = []
for content_image, style_image in zip(content_images, style_images):
    results_path = run_neural_style_transfer(content_image, style_image)
    result_paths.append(results_path)

/usr/local/lib/python3.10/dist-packages/torchvision/models/_utils.py:208: UserWarning: The parameter 'pretrained' is deprecated since 0.13 and may be removed in the future, please use 'weights' instead.
  warnings.warn(
/usr/local/lib/python3.10/dist-packages/torchvision/models/_utils.py:223: UserWarning: Arguments other than a weight enum or `None` for 'weights' are deprecated since 0.13 and may be removed in the future. The current behavior is equivalent to passing `weights=VGG19_Weights.IMAGENET1K_V1`. You can also use `weights=VGG19_Weights.DEFAULT` to get the most up-to-date weights.
  warnings.warn(msg)

Using VGG19 in the optimization procedure.
L-BFGS | iteration: 000, total loss=3446674817024.0000, content_loss=      0.0000, style loss=3446667600000.0000, tv loss=7292746.0000
L-BFGS | iteration: 100, total loss=36545269760.0000, content_loss=1636035644.5312, style loss=34891616250.0000, tv loss=17615628.0000
L-BFGS | iteration: 200, total loss=11921997824.0000, content_loss=1652365625.0000, style loss=10250455312.5000, tv loss=19177282.0000
L-BFGS | iteration: 300, total loss=7036205568.0000, content_loss=1648760351.5625, style loss=5368206562.5000, tv loss=19238864.0000
L-BFGS | iteration: 400, total loss=5205744128.0000, content_loss=1640404687.5000, style loss=3546391171.8750, tv loss=18947876.0000
L-BFGS | iteration: 500, total loss=4413362176.0000, content_loss=1627640234.3750, style loss=2767006406.2500, tv loss=18715636.0000
L-BFGS | iteration: 600, total loss=3994198528.0000, content_loss=1615958691.4062, style loss=2359643203.1250, tv loss=18596980.0000
L-BFGS | iteration: 700, total loss=3734805760.0000, content_loss=1603115527.3438, style loss=2113161562.5000, tv loss=18528570.0000
L-BFGS | iteration: 800, total loss=3555087872.0000, content_loss=1591160742.1875, style loss=1945416210.9375, tv loss=18510824.0000
L-BFGS | iteration: 900, total loss=3416918272.0000, content_loss=1581310839.8438, style loss=1817070000.0000, tv loss=18537596.0000
L-BFGS | iteration: 1000, total loss=3312789760.0000, content_loss=1572570996.0938, style loss=1721618437.5000, tv loss=18600136.0000
Using VGG19 in the optimization procedure.
L-BFGS | iteration: 000, total loss=11997666082816.0000, content_loss=      0.0000, style loss=11997665280000.0000, tv loss=1385193.0000
L-BFGS | iteration: 100, total loss=13115474944.0000, content_loss=1575101367.1875, style loss=11502725625.0000, tv loss=37647888.0000
L-BFGS | iteration: 200, total loss=4672230400.0000, content_loss=1514814550.7812, style loss=3115721250.0000, tv loss=41694832.0000
L-BFGS | iteration: 300, total loss=3249261056.0000, content_loss=1465064746.0938, style loss=1740253476.5625, tv loss=43943024.0000
L-BFGS | iteration: 400, total loss=2703843328.0000, content_loss=1425714160.1562, style loss=1232585507.8125, tv loss=45543648.0000
L-BFGS | iteration: 500, total loss=2434893568.0000, content_loss=1393794824.2188, style loss=994457226.5625, tv loss=46641420.0000
L-BFGS | iteration: 600, total loss=2279518720.0000, content_loss=1368256933.5938, style loss=863873144.5312, tv loss=47388748.0000
L-BFGS | iteration: 700, total loss=2176390912.0000, content_loss=1346918750.0000, style loss=781506035.1562, tv loss=47966200.0000
L-BFGS | iteration: 800, total loss=2103154048.0000, content_loss=1330168750.0000, style loss=724612910.1562, tv loss=48372392.0000
L-BFGS | iteration: 900, total loss=2051134976.0000, content_loss=1317141308.5938, style loss=685334824.2188, tv loss=48658900.0000
L-BFGS | iteration: 1000, total loss=2009840768.0000, content_loss=1305658886.7188, style loss=655317714.8438, tv loss=48864164.0000
Using VGG19 in the optimization procedure.
L-BFGS | iteration: 000, total loss=5273057492992.0000, content_loss=      0.0000, style loss=5273050560000.0000, tv loss=6960354.0000
Using VGG19 in the optimization procedure.
L-BFGS | iteration: 000, total loss=444431302656.0000, content_loss=      0.0000, style loss=444426870000.0000, tv loss=4424529.0000
L-BFGS | iteration: 100, total loss=1352650624.0000, content_loss=772381738.2812, style loss=557559960.9375, tv loss=22708902.0000
L-BFGS | iteration: 200, total loss=927935168.0000, content_loss=666212988.2812, style loss=240798750.0000, tv loss=20923466.0000
L-BFGS | iteration: 300, total loss=824836736.0000, content_loss=624409570.3125, style loss=181657368.1641, tv loss=18769770.0000
L-BFGS | iteration: 400, total loss=773977408.0000, content_loss=600071777.3438, style loss=157143090.8203, tv loss=16762471.0000
L-BFGS | iteration: 500, total loss=746054528.0000, content_loss=585927148.4375, style loss=144878466.7969, tv loss=15248924.0000
L-BFGS | iteration: 600, total loss=729051136.0000, content_loss=576800195.3125, style loss=137999311.5234, tv loss=14251665.0000
L-BFGS | iteration: 700, total loss=717764672.0000, content_loss=570511523.4375, style loss=133622402.3438, tv loss=13630727.0000
L-BFGS | iteration: 800, total loss=710304064.0000, content_loss=565828759.7656, style loss=131205629.8828, tv loss=13269675.0000
L-BFGS | iteration: 900, total loss=705184960.0000, content_loss=562691894.5312, style loss=129458715.8203, tv loss=13034339.0000
L-BFGS | iteration: 1000, total loss=701642816.0000, content_loss=560543408.2031, style loss=128230063.4766, tv loss=12869339.0000
Using VGG19 in the optimization procedure.
L-BFGS | iteration: 000, total loss=278622928896.0000, content_loss=      0.0000, style loss=278611650000.0000, tv loss=11262601.0000
L-BFGS | iteration: 100, total loss=5382811136.0000, content_loss=576838818.3594, style loss=4792199062.5000, tv loss=13773154.0000
L-BFGS | iteration: 200, total loss=1557859712.0000, content_loss=557504394.5312, style loss=985689375.0000, tv loss=14665967.0000
L-BFGS | iteration: 300, total loss=1022564608.0000, content_loss=521756054.6875, style loss=486433740.2344, tv loss=14374810.0000
L-BFGS | iteration: 400, total loss=816369792.0000, content_loss=498498925.7812, style loss=303496787.1094, tv loss=14374097.0000
L-BFGS | iteration: 500, total loss=736841792.0000, content_loss=484446386.7188, style loss=238206679.6875, tv loss=14188733.0000
L-BFGS | iteration: 600, total loss=698386688.0000, content_loss=474832470.7031, style loss=209693159.1797, tv loss=13861061.0000
L-BFGS | iteration: 700, total loss=675974400.0000, content_loss=468174316.4062, style loss=194242895.5078, tv loss=13557195.0000
L-BFGS | iteration: 800, total loss=660216640.0000, content_loss=463156689.4531, style loss=183744448.2422, tv loss=13315493.0000
L-BFGS | iteration: 900, total loss=649254400.0000, content_loss=459336767.5781, style loss=176767836.9141, tv loss=13149842.0000
L-BFGS | iteration: 1000, total loss=640705152.0000, content_loss=456254248.0469, style loss=171415356.4453, tv loss=13035504.0000
Using VGG19 in the optimization procedure.
L-BFGS | iteration: 000, total loss=903369261056.0000, content_loss=      0.0000, style loss=903366240000.0000, tv loss=2991186.0000
L-BFGS | iteration: 100, total loss=29200771072.0000, content_loss=885201464.8438, style loss=28309640625.0000, tv loss=5929405.5000
L-BFGS | iteration: 200, total loss=6572710400.0000, content_loss=998418847.6562, style loss=5566220625.0000, tv loss=8070559.0000
L-BFGS | iteration: 300, total loss=3120206336.0000, content_loss=984832226.5625, style loss=2127471328.1250, tv loss=7902820.5000
L-BFGS | iteration: 400, total loss=2409276928.0000, content_loss=969420019.5312, style loss=1431975234.3750, tv loss=7881726.0000
L-BFGS | iteration: 500, total loss=2078447104.0000, content_loss=955389648.4375, style loss=1115086640.6250, tv loss=7970847.0000
L-BFGS | iteration: 600, total loss=1908204672.0000, content_loss=941749804.6875, style loss=958418730.4688, tv loss=8036112.0000
L-BFGS | iteration: 700, total loss=1813783808.0000, content_loss=931931835.9375, style loss=873780937.5000, tv loss=8071047.0000
L-BFGS | iteration: 800, total loss=1754478208.0000, content_loss=923950000.0000, style loss=822423867.1875, tv loss=8104305.0000
L-BFGS | iteration: 900, total loss=1713480320.0000, content_loss=918448925.7812, style loss=786893085.9375, tv loss=8138319.0000
L-BFGS | iteration: 1000, total loss=1682264448.0000, content_loss=913685058.5938, style loss=760401738.2812, tv loss=8177604.0000


for content_image, style_image, results_path in zip(
    content_images, style_images, result_paths
):
    visualize_style_transfer(content_image, style_image, results_path)


# подключим Google Drive
from google.colab import drive

drive.mount("/content/drive")

# пути к нужным папкам в Drive
content_path = "/content/drive/MyDrive/DS/style_transfer_images/content"
style_path = "/content/drive/MyDrive/DS/style_transfer_images/style"

# очистим предыдущие картинки, скопируем новые
!rm -r data/content-images
!mkdir data/content-images
!cp -a /$content_path/. data/content-images/

!rm -r data/style-images
!mkdir data/style-images
!cp -a /$style_path/. data/style-images/

Drive already mounted at /content/drive; to attempt to forcibly remount, call drive.mount("/content/drive", force_remount=True).


content_images = sorted(os.listdir("data/content-images"))
style_images = sorted(os.listdir("data/style-images"))
result_paths = []
for content_image, style_image in zip(content_images, style_images):
    results_path = run_neural_style_transfer(content_image, style_image)
    result_paths.append(results_path)

Using VGG19 in the optimization procedure.
L-BFGS | iteration: 000, total loss=951764320256.0000, content_loss=      0.0000, style loss=951746580000.0000, tv loss=17784880.0000
L-BFGS | iteration: 100, total loss=3000761088.0000, content_loss=1021630273.4375, style loss=1939623750.0000, tv loss=39506920.0000
L-BFGS | iteration: 200, total loss=1648944896.0000, content_loss=925682128.9062, style loss=682582675.7812, tv loss=40680200.0000
L-BFGS | iteration: 300, total loss=1358226688.0000, content_loss=863602539.0625, style loss=453543896.4844, tv loss=41080144.0000
L-BFGS | iteration: 400, total loss=1244255872.0000, content_loss=828277636.7188, style loss=374861865.2344, tv loss=41116412.0000
L-BFGS | iteration: 500, total loss=1182209920.0000, content_loss=804843310.5469, style loss=336351298.8281, tv loss=41015240.0000
L-BFGS | iteration: 600, total loss=1143759104.0000, content_loss=788581005.8594, style loss=314325322.2656, tv loss=40852712.0000
L-BFGS | iteration: 700, total loss=1116899712.0000, content_loss=776487841.7969, style loss=299742978.5156, tv loss=40668804.0000
L-BFGS | iteration: 800, total loss=1097843840.0000, content_loss=768284619.1406, style loss=289046689.4531, tv loss=40512528.0000
L-BFGS | iteration: 900, total loss=1083036928.0000, content_loss=761707763.6719, style loss=280974755.8594, tv loss=40354380.0000
L-BFGS | iteration: 1000, total loss=1071727040.0000, content_loss=756704345.7031, style loss=274824814.4531, tv loss=40197940.0000
Using VGG19 in the optimization procedure.
L-BFGS | iteration: 000, total loss=2997639446528.0000, content_loss=      0.0000, style loss=2997616080000.0000, tv loss=23310342.0000
L-BFGS | iteration: 100, total loss=7115043840.0000, content_loss=2474382031.2500, style loss=4605689062.5000, tv loss=34972816.0000
L-BFGS | iteration: 200, total loss=4868994048.0000, content_loss=2367574804.6875, style loss=2465299453.1250, tv loss=36119584.0000
L-BFGS | iteration: 300, total loss=4269140992.0000, content_loss=2298016796.8750, style loss=1934330976.5625, tv loss=36793292.0000
L-BFGS | iteration: 400, total loss=3988312576.0000, content_loss=2251633007.8125, style loss=1699445507.8125, tv loss=37234076.0000
L-BFGS | iteration: 500, total loss=3824046336.0000, content_loss=2219921484.3750, style loss=1566597421.8750, tv loss=37527608.0000
L-BFGS | iteration: 600, total loss=3718422016.0000, content_loss=2196760742.1875, style loss=1483916484.3750, tv loss=37744720.0000
L-BFGS | iteration: 700, total loss=3643614720.0000, content_loss=2178483398.4375, style loss=1427200546.8750, tv loss=37930768.0000
L-BFGS | iteration: 800, total loss=3588964608.0000, content_loss=2164393359.3750, style loss=1386471796.8750, tv loss=38099224.0000
L-BFGS | iteration: 900, total loss=3545565440.0000, content_loss=2153188281.2500, style loss=1354139648.4375, tv loss=38237388.0000
L-BFGS | iteration: 1000, total loss=3511093504.0000, content_loss=2144098828.1250, style loss=1328638359.3750, tv loss=38356236.0000
Using VGG19 in the optimization procedure.
L-BFGS | iteration: 000, total loss=4883807731712.0000, content_loss=      0.0000, style loss=4883785920000.0000, tv loss=22038304.0000
L-BFGS | iteration: 100, total loss=12774153216.0000, content_loss=2611515820.3125, style loss=10122080625.0000, tv loss=40557988.0000
L-BFGS | iteration: 200, total loss=6340702720.0000, content_loss=2525504882.8125, style loss=3772573125.0000, tv loss=42624376.0000
L-BFGS | iteration: 300, total loss=4932108288.0000, content_loss=2441932812.5000, style loss=2446583437.5000, tv loss=43591816.0000
L-BFGS | iteration: 400, total loss=4326545920.0000, content_loss=2380906640.6250, style loss=1901389921.8750, tv loss=44249336.0000
L-BFGS | iteration: 500, total loss=3979491328.0000, content_loss=2335890039.0625, style loss=1598844023.4375, tv loss=44757376.0000
L-BFGS | iteration: 600, total loss=3764689920.0000, content_loss=2300800000.0000, style loss=1418788359.3750, tv loss=45101616.0000
L-BFGS | iteration: 700, total loss=3615907584.0000, content_loss=2273895507.8125, style loss=1296713085.9375, tv loss=45299000.0000
L-BFGS | iteration: 800, total loss=3508451584.0000, content_loss=2250956445.3125, style loss=1212083320.3125, tv loss=45411536.0000
L-BFGS | iteration: 900, total loss=3428905984.0000, content_loss=2232491210.9375, style loss=1150955039.0625, tv loss=45459740.0000
L-BFGS | iteration: 1000, total loss=3366068224.0000, content_loss=2217284570.3125, style loss=1103314687.5000, tv loss=45468892.0000
Using VGG19 in the optimization procedure.
L-BFGS | iteration: 000, total loss=5146914848768.0000, content_loss=      0.0000, style loss=5146897440000.0000, tv loss=17441590.0000
L-BFGS | iteration: 100, total loss=97621319680.0000, content_loss=3216366796.8750, style loss=94376392500.0000, tv loss=28560512.0000
L-BFGS | iteration: 200, total loss=31281483776.0000, content_loss=3256056835.9375, style loss=27996470625.0000, tv loss=28953982.0000
L-BFGS | iteration: 300, total loss=14656659456.0000, content_loss=3268785546.8750, style loss=11359070625.0000, tv loss=28802968.0000
L-BFGS | iteration: 400, total loss=9275234304.0000, content_loss=3267282226.5625, style loss=5979250312.5000, tv loss=28701944.0000
L-BFGS | iteration: 500, total loss=7155175424.0000, content_loss=3250802929.6875, style loss=3875898750.0000, tv loss=28473836.0000
L-BFGS | iteration: 600, total loss=6092583424.0000, content_loss=3238119726.5625, style loss=2826218671.8750, tv loss=28244948.0000
L-BFGS | iteration: 700, total loss=5451201024.0000, content_loss=3222942773.4375, style loss=2200143515.6250, tv loss=28114406.0000
L-BFGS | iteration: 800, total loss=5025047040.0000, content_loss=3207201757.8125, style loss=1789825312.5000, tv loss=28020074.0000
L-BFGS | iteration: 900, total loss=4729344000.0000, content_loss=3193596679.6875, style loss=1507825781.2500, tv loss=27921448.0000
L-BFGS | iteration: 1000, total loss=4517608960.0000, content_loss=3180410546.8750, style loss=1309397460.9375, tv loss=27801096.0000


# Сохраним результаты
!cp -r data/output-images /content/drive/MyDrive/DS

# Визуализируем
for content_image, style_image, results_path in zip(
    content_images, style_images, result_paths
):
    visualize_style_transfer(content_image, style_image, results_path)


!pip install -q diffusers

     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 1.9/1.9 MB 14.1 MB/s eta 0:00:00


from diffusers import AutoPipelineForText2Image
import torch

pipe = AutoPipelineForText2Image.from_pretrained(
    "stabilityai/sdxl-turbo", torch_dtype=torch.float16, variant="fp16"
).to("cuda")

The cache for model files in Transformers v4.22.0 has been updated. Migrating your old cache. This is a one-time only operation. You can interrupt this and resume the migration later on by calling `transformers.utils.move_cache()`.

0it [00:00, ?it/s]

/usr/local/lib/python3.10/dist-packages/huggingface_hub/utils/_token.py:88: UserWarning: 
The secret `HF_TOKEN` does not exist in your Colab secrets.
To authenticate with the Hugging Face Hub, create a token in your settings tab (https://huggingface.co/settings/tokens), set it as secret in your Google Colab and restart your session.
You will be able to reuse this secret in all of your notebooks.
Please note that authentication is recommended but still optional to access public models or datasets.
  warnings.warn(

model_index.json:   0%|          | 0.00/685 [00:00<?, ?B/s]

Fetching 18 files:   0%|          | 0/18 [00:00<?, ?it/s]

scheduler/scheduler_config.json:   0%|          | 0.00/459 [00:00<?, ?B/s]

text_encoder_2/config.json:   0%|          | 0.00/575 [00:00<?, ?B/s]

text_encoder/config.json:   0%|          | 0.00/565 [00:00<?, ?B/s]

tokenizer/tokenizer_config.json:   0%|          | 0.00/704 [00:00<?, ?B/s]

tokenizer/special_tokens_map.json:   0%|          | 0.00/586 [00:00<?, ?B/s]


prompt = "a photo of a parrot riding a scooter in Moscow, artstation, highly detailed"

# Передаем prompt в pipe, указываем параметры
# Модель проводит генерацию картинки, соответствующей запросу
image = pipe(prompt=prompt, num_inference_steps=1, guidance_scale=0.0).images[0]
image

  0%|          | 0/1 [00:00<?, ?it/s]


prompt = "computer science student pulling an all-nighter, artstation, extremely detailed"
image = pipe(prompt=prompt, num_inference_steps=1, guidance_scale=0.0).images[0]
image

  0%|          | 0/1 [00:00<?, ?it/s]


prompt = "Valparaiso city view on a sunny day with a rainbow, painting, artstation, extremely detailed"
image = pipe(prompt=prompt, num_inference_steps=1, guidance_scale=0.0).images[0]
image

  0%|          | 0/1 [00:00<?, ?it/s]


prompt = "cat as the ultimate autonomous entity with sacred geometry, artstation, highly detailed"
image = pipe(prompt=prompt, num_inference_steps=1, guidance_scale=0.0).images[0]
image

  0%|          | 0/1 [00:00<?, ?it/s]

Введение в анализ данных ¶

Компьютерное зрение & генеративные модели¶

2. Нейросетевой перенос стиля¶

3. Диффузионные модели¶

Введение в анализ данных¶

Компьютерное зрение & генеративные модели¶

2. Нейросетевой перенос стиля¶

3. Диффузионные модели¶

Введение в анализ данных ¶